Running Experiments on GrowthBook
GrowthBook has a few different ways to run experiments or AB tests depending on your needs. This guide will walk you through the different ways of running experiments on GrowthBook.
Server Side and Mobile Experiments
Server-side A/B testing, also known as backend or server-side experimentation, is a technique used in software development and web applications to test measure the impact of any code changes or new features. The changes may impact both the user interface and the backend of the application, but the decision about what version to serve a user is decided on the server. This has a number of advantages over client side experiments, specifically, that it allows you to run very complex tests that may involve a lot of different parts of the code, and span multiple parts of your application. It also avoids any issues with flickering that can happen with client side testing.
Feature Flag Experiments
With GrowthBook, the easiest way to do server-side testing is by using feature flags. Each feature flag has conditional logic (rules) which controls how a feature should be shown, and if a feature should be shown as part of an experiment. GrowthBook also lets you target any feature or rule based on the targeting attributes you define. With GrowthBook, you can add an experiment rule to a feature that will randomly assign the users based on some hashing attribute into one of your experiment variations. You can read more about feature flag experiment rules here, or more details on running an experiment with feature flags.
You can also run server-side experiments by using inline experiments directly with our SDK. This requires no 3rd party requests, as the experiment conditions and settings can be written directly into the code. How you implement in-line experiments depends on the SDK language you're using, you can read more about this from our SDKs pages.
Client Side (Browser) Experiments
When the client loads your application, our SDK will check to see if the user should be part of any experiments, and if so, assign them and the client code will serve that variations treatment to them. GrowthBook's client side SDKs can handle all of this for you, and you can also serve client-side A/B test via our feature flags, or using our visual editor.
Client-side A/B tests are great for testing visual changes to your application, but they do have some drawbacks. One of the most common issues is caused by the delay in loading the specific variation to a user, which may cause a flash or flickering as the experiment loads. This can be reduced by using inline-experiments, or by moving the GrowthBook SDK code higher in the code so it loads at the same time or before the page. Sometimes the hashing attribute may not be available until your tracking software has loaded, and you may need to use a custom tracking id.
GrowthBook's feature flags can work just as well from the client/browser as it does server side. The feature flag's conditional logic (rules) controls how a feature should be shown, and if a feature should be shown as part of an experiment. The same targeting and override rules apply exactly the same to client-side experimentation.
Just like with inline server-side experiments, you can run experiments inline on the client. This requires no 3rd party requests, as the experiment conditions and settings can be written directly into the code, and processed by the SDK. How you implement in-line experiments depends on the SDK language you're using, you can read more about this from our SDKs pages.
Visual Editor (WYSIWYG) Experiments
GrowthBook has a visual editor for running experiments on the front-end of your website without requiring any code changes. Our visual editor uses the same client-side SDK used for feature flags and A/B testing.
API / ML Experiments
GrowthBook's SDKs works well with anywhere code can run, and as such you can use it in the API or when running machine learning models. And, with our deterministic hashing method for assignment, you can even be sure users get assigned the same variation across your platform without needing to store state from GrowthBook.
Custom Assignment or 3rd Party Experiments
GrowthBook is a modular platform and can be used for experiment analysis if you are using custom assignment code or another experimentation system for randomization of users into variations. As long as the exposure/assignment information is available from within your data warehouse, you can use GrowthBook to analyse the results of your experiments.
How GrowthBook Assigns Users to Experiments
GrowthBook uses a consistent hashing algorithm to assign users to experiments. This means that the same user will always get the same variation given the same experiment settings (experiment key, and user hashing id). This is useful because it means that you can run experiments across multiple pages, or even multiple applications, and the user will always get the same variation. It also means that GrowthBook stores no state, and requires no additional cookies. It's important to note that none of these implies that GrowthBook exposes a mechanism known as sticky-bucketing. It's all about the consistent evaluation of variations based on the particular experiment settings.
This is a subset of some of the information you can find about running experiments in our Open Guide to AB Testing.
Running A/A Tests
An A/A test is the same as an A/B test, but each variation has no actual difference in the application. This lets you test out that your systems are working correctly, as you should see no significant differences between the variations. We suggest that you first an A/A test to validate your experimentation implementation is correctly splitting traffic, and producing statistically valid results.
When to Expose Users to Experiments
When running an experiment it is best if you can only expose users as close to the actual treatment exposure as possible. This means that if you're testing something like new signup flow, you don't expose all users, including those who never open that window. Including users who did not see the treatment will increase the noise and reduce the ability to detect any differences.
If assignment is unavoidably separated from exposure, you can use an activation metric to filter out these un-exposed users from the analysis.
Flickering with front end or client-side A/B tests is an artifact from all client-side A/B testing tools. This is caused by a delay in loading the specific variation for a user, which may cause a flash or flickering as the experiment loads. This can be reduced by using inline-experiments, or by moving the GrowthBook SDK code higher in the code file, or using server side A/B tests.
There are a few other ways to reduce flickering that some platforms utilize. One common "flicker free" technique is to load a white overlay, or just hide parts of the page, as the page loads. The result is that users cannot see any flickering that may be happening beneath the overlay as the variations are loaded.
Understanding experiment power and MDE are important to predict how many samples are required. There are numerous online calculators that can be used to help you predict the sample size. Typical rule of thumb for the lowest number of samples required is that you want at least 200 conversion events per variation. So for example if you have a registration page which has a 10% conversion rate, and you have a 2 way (A and B) experiment that is looking to improve the member registrations, you will want to expose the experiment to at least 4,000 people (2000 per variation).
Due to the natural variability in traffic day to day and hour to hour, experimentation teams will often set a minimum test duration within which a test cannot be called. This helps you avoid optimizing a product for just the users that happen to visit when the test is started. For example, if the weekend traffic of your product is different from the traffic during the week, if you started a test on Friday and ended it on Monday, you may not get a complete picture of the impact your changes have to your weekday traffic.
Typical test durations are 1 to 2 weeks, and usually care needs to be taken over holidays. You may also find that a test would need to run for a month or more to get the power required for the experiment. Very long running tests can be hard to justify as you have to keep the variations of the experiment unchanged for duration, and this may limit your team's velocity towards potentially higher impact changes.
Interaction Effects and Mutual Exclusion
When you start having the ability to run a lot of A/B tests, you may start worrying about how tests running in parallel may interact and effect the other results. For example you may want to test a change in the CTA button on your purchase page, and also test changing the price. It can be difficult to figure out if any two tests will meaningfully interact, and many will run the tests in serial in an abundance of caution.
However, meaningful interactions are actually quite rare, and keeping a higher rate of experimentation is usually more beneficial. You can run analysis after the experiments to see if there were any interaction effects which would change your conclusions. If you need to run mutually exclusive tests, you can use GrowthBook’s namespace feature.
Having a high frequency of A/B testing is important for running a success experimentation program. The main reasons why experimentation frequency is important are:
- Maximizing chances: Since success rates are typically low for any given experiment, and large changes are even more rare, by having a high frequency of A/B testing you are maximizing your chance of having impactful experiments.
- Continuous improvement: A high frequency of A/B testing allows you to continuously improve your website or application. By testing small changes frequently, you can quickly identify and implement changes that improve user experience, engagement, and conversion rates.
- Adaptability: A high frequency of A/B testing allows you to quickly adapt to changes in user behavior, market trends, or other external factors that may impact your website or application. By testing frequently, you can identify and respond to these changes more quickly, ensuring that your site or app remains relevant and effective.
- Avoiding stagnation: A high frequency of A/B testing can help you avoid stagnation and complacency. By continually testing and experimenting, you can avoid falling into a rut or becoming overly attached to a specific design or strategy, and instead remain open to new ideas and approaches.